EDITORIAL ARCHIVE
AI news
News, tests and analysis to understand what changed, what is hype and what your business can already use.
Latest from the radar
Page 4 / 25CONFIRMED: OpenAI Brings GPT-Live-1 to Developers at $0.05 per Minute
OpenAI released GPT-Live-1 in the API on September 10, 2026, with full duplex voice, telephony support, and a $0.05 per minute price for the voice layer. The move turns a developer promise into usable infrastructure for real time service agents and operations.
CONFIRMED: NSA, CISA and FBI accuse DeepSeek, Alibaba and four more Chinese firms of industrial scale distillation of Claude, GPT, Gemini and Grok
Label: CONFIRMED. A joint bulletin from the NSA, CISA and FBI, published September 8, 2026, accuses DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI of systematically extracting capabilities from American models such as Claude, GPT, Gemini and Grok since late 2024, likely with the Chinese government's knowledge. The document names specific source models and describes industrial scale evasion tactics.
CONFIRMED: DeepSeek launches V4.1 Flash with native vision and cuts API prices
DeepSeek has confirmed the launch of V4.1 Flash, the smallest model in its new architecture family, with native visual understanding, a 1 million token context, tool support, and an MoE architecture designed to reduce memory costs. The lab also announced that V4 Pro requests will be routed to Flash from September 14 until V4.1 Pro arrives.
STRONG SIGNAL: Alexandr Wang confirms Watermelon, Meta's next frontier model codename
Days after shipping Muse Spark 1.3 and claiming parity with Fable 5.1 and Opus 5 on independent benchmarks, Meta's Chief AI Officer Alexandr Wang publicly confirmed the company is building a much larger model called Watermelon. Internal documents cited by the press point to an October target, but Meta has not confirmed a date, specs, or the consumer agent platform Hatch reportedly launching alongside it.
STRONG SIGNAL: code from three inference engines already runs "qwen4_exp", real clue that Qwen4 is leaving the drawing board
Label: STRONG SIGNAL. An open pull request in vLLM, another in SGLang, and commits in two independent Apple Silicon serving projects already treat an identifier called qwen4_exp as a real architecture, with dedicated kernels. Alibaba has never announced a date, price, or spec for Qwen4, but the engineering trail is convergent and verifiable in public code.
Claude Code 2.1.266 Fixes a Regression That Broke Gateways and Proxies
CONFIRMED. Anthropic released Claude Code 2.1.266 to undo a regression introduced in the previous version: CLAUDE_CODE_USE_GATEWAY is ignored again when used on its own, preventing configurations with an API key, apiKeyHelper, or custom headers from being forced into Cloud gateway sign-in.
GPT-5.6 Sol takes over quantum-chip measurements and shows the next step for agents
OpenAI showed GPT-5.6 Sol connected to Codex and laboratory software to run routine measurements on quantum chips. The case points to agents operating physical workflows while exposing their current limits with weak and ambiguous signals.
Meta introduces Muse, a personal agent built to act on your behalf
Meta introduced Muse, a personal agent that browses, fills out forms and keeps working after users close the app. The rollout starts in the United States, shifting the race toward reliable execution rather than better answers.
GPT Image 2.5 gets official support in OpenAI’s Python SDK
CONFIRMED: version 3.10.0 of OpenAI’s Python SDK added GPT Image 2.5 models and image options, turning the visual launch into an official integration surface for applications. The same version added expiration fields for service-account API keys.
RUMOR: DeepSeek V5 reportedly built from scratch to rival Claude Mythos
A post on X from leak tracker Mr_Salio, circulated on September 7, 2026, claims DeepSeek is building V5 from scratch, targeting Anthropic's Mythos tier and OpenAI's GPT-6 Astra, while running cheaper than both. No model card, benchmark, or API entry exists. DeepSeek has never confirmed the name.
Mistral Raises €3 Billion, Turning European Sovereignty Into a Scale Thesis
CONFIRMED. Mistral announced a €3 billion Series D led by Samsung, at a post money valuation above €21 billion. The capital will expand frontier research, compute capacity and international growth, but it is not yet evidence of a product, pricing or performance breakthrough.
iFlytek launches Spark X2.5, a 293B model trained end to end on Chinese chips
iFlytek officially launched Spark X2.5 this Monday, a 293 billion parameter MoE model focused on coding and agents, trained and served entirely on domestic Chinese compute. The launch reinforces China's push for chip sovereignty, but has no independent benchmark yet against Qwen3, DeepSeek or GLM 5.